A charged particle beam lithography system such as the electron beam lithography system or the ion beam lithography system focuses the charged particle beam from the charged particle beam generating means onto the material on which patterns are to be written. The material is scanned with said charged particle beam by using a deflection system. Scanning is based on the size of each pattern to be written and the computer stored data corresponding to the writing position. Thus, each specified pattern is written on the material.
Optical devices that are formed by charged particle beam lithography include those that have grating patterns with various pitches. These optical devices are designed to be irradiated with various wavelengths to analyze the irradiating light. In fabricating these optical devices, it may be desired to reduce or expand the size of the grating pattern obtained by lithography in order to set the resolution of light at the desired value. In such a case, the conventional art multiplies the computer stored data by a scaling factor when converting the data from the pattern source data (design data) into writing data that can be handled by the lithography system. Otherwise, the computer stored source data is redesigned according to the scaling factor. Such a scaling is disclosed in U.S. Pat. No. 4,530,064.
In a lithography system wherein the number of bits of the digital circuit for converting the deflection signal into an analog signal is finite, there exists a minimum value of data increment. Thus, the former method which multiplies the data by a scaling factor when converting the source data into the writing data cannot change the pattern pitch data to a value smaller than the minimum value of the aforementioned data increment and the data is rounded by this minimum value. For example, when the data representing pitch between the grating patterns corresponds 100 nm and the minimum increment value is 5 nm, the pitch data becomes 95 nm, even if the pitch data is multiplied by the scaling factor 0.99. On the other hand, the latter method has a problem that the source data must be redesigned according to the scaling factor, which requires much computer processing time.